We will apply Dirichlet regression with Braak stage as the primary predictor to SEA-AD snRNA-seq subtype proportions (84 donors, DLPFC) and replicate in ROSMAP (Mathys 2023, ~427 donors). For each major microglia subtype (homeostatic MG-1, transitional MG-2, DAM/MGnD MG-3, MG-4) and each oligodendrocyte/OPC subtype, we will estimate Braak-stage inflection points using piecewise-linear or segmented regression with bootstrap 95% CIs. Cross-lineage comparison will test whether the microglia inflection precedes, follows, or overlaps the oligo/OPC inflection (non-overlapping CIs as the primary criterion). Sex, APOE genotype, and post-mortem interval will enter as covariates. Replication in ROSMAP and directional consistency with Green et al. 2024 cellular communities will be required before causal language is used.
Details
- disease
- alzheimer's disease
- target_ref
- dataset:SEA-AD-snRNA-DLPFC
- identification_strategy
- observational
Raw fields (4)
- assay_spec
Primary cohort: SEA-AD snRNA-seq DLPFC, 84 donors, accessed via CELLxGENE Census Python API (cellxgene_census.get_obs, tissue_general='prefrontal cortex', dataset_title contains 'SEA-AD'). Microglia subtypes resolved from obs field 'subclass_label' (homeostatic: P2RY12+/TMEM119+; DAM/MGnD: SPP1+/GPNMB+/TREM2+; IRM: IFIT1+/IFIT3+; LDA: LPL+/PLIN2+). OPC and oligodendrocyte subtypes from same field (OPC: PDGFRA+; mature oligo: MBP+/MOG+). Predictor: braak_int (0–6, mapped from Braak_stage obs field). Covariates: sex, PMI, RIN, age_at_death. Replication cohort: Mathys 2023 Cell (DOI: 10.1016/j.cell.2023.08.039), GEO GSE174367, per-donor proportion table from supplement. Independent validation: Green 2024 Nature (DOI: 10.1038/s41586-024-07871-6) ROSMAP cellular-community fractions. Statistical model: scCODA (pertpy>=0.7 Sccoda), num_samples=10000, num_warmup=2000, credible_effects_threshold=0.95. Planned output: inflection-point Braak stage per subtype (posterior effect sign reversal or credible threshold crossing), reported with 95% HDI.
- hypothesis
DAM/MGnD microglia (SPP1+, GPNMB+, TREM2+) proportions rise at an earlier Braak stage than oligodendrocyte/OPC proportions fall, indicating that neuroinflammatory expansion precedes—and may drive—oligodendroglial depletion in AD cortex rather than the reverse.
- kill_criteria
Abort or revise if: (1) CELLxGENE Census obs lacks donor_id or Braak_stage — BLOCKED, file Allen portal data-request; (2) R-hat > 1.05 on any posterior parameter in microglia or oligo model — flag, withhold interpretation; (3) Mathys 2023 GEO supplement lacks per-donor count matrix with Braak metadata — replication arm BLOCKED, fall back to Mathys 2019 or report single-cohort only; (4) fewer than 5 donors per Braak stage bin in SEA-AD — compositional model underpowered, report with caveat; (5) Green 2024 community fractions do not map to microglia or oligo subtypes — validation arm BLOCKED; (6) logistic curve_fit fails to converge for >50% of subtypes — inflection detection BLOCKED, use empirical mode instead; (7) DAM/MGnD posterior HDI overlaps OPC/oligo nadir HDI across all Braak bins — hypothesis NOT supported, revise to null-result artifact.
- primary_endpoint
Tick 12: Targeted EuropePMC sweeps with shorter query strings — (1) microglia DAM Braak snRNA human, (2) OPC/oligodendrocyte depletion AD cortex snRNA — plus CrossRef provenance stamps for Mathys 2019 (DOI 10.1038/s41586-019-1195-2) and Mathys 2023 (DOI 10.1016/j.cell.2023.08.039) as ROSMAP anchor references. Success criterion: ≥1 hit per sweep with human snRNA-seq microglia subtype or OPC staging data; if both criteria met, advance to notebook scaffold tick.